[Bug other/16796] New: PowerPC - Unnecessary Floating Point Register Copy

[Bug other/16796] New: PowerPC - Unnecessary Floating Point Register Copy

[gcc-bugzilla at gcc dot gnu dot org]

Description:
A non-optimal code sequence is illustraded.  A floating point register copy is unnecessary and could be eliminated.  Duplicate using gcc 3.5 and command line:

gcc -O3 -m64 -c test.c

Testcase:
static const double huge = 1.0e300;
typedef int int64_t __attribute__ ((__mode__ (__DI__)));
typedef unsigned int u_int64_t __attribute__ ((__mode__ (__DI__)));

double __floor(double x)
{
  union {
    double dbl_val;
    long int long_val;
  } temp;

  int64_t i0,j0;
  u_int64_t i;
  temp.dbl_val = x;
  i0 = temp.long_val;

  j0 = ((i0>>52)&0x7ff)-0x3ff;
  if(j0<52) {
    if(j0<0) {
      if(huge+x>0.0) {
        if(i0>=0) {i0=0;}
        else if((i0&0x7fffffffffffffff)!=0)
        { i0=0xbff0000000000000;}
      }
    } else {
      i = (0x000fffffffffffff)>>j0;
      if((i0&i)==0) return x;
      if(huge+x>0.0) {
        if(i0<0) i0 += (0x0010000000000000)>>j0;
        i0 &= (~i);
      }
    }
  } else {
    if (j0==0x400)
      return x+x;
    else
      return x;
  }
  temp.long_val = i0;
  x = temp.dbl_val;
  return x;
}

Assembly:
.__floor:
	stfd 1,-16(1)
	ld 0,-16(1)
	rldicl 9,0,12,53
	mr 11,0
	addi 9,9,-1023
	cmpdi 7,9,51
	bgt- 7,.L2
	cmpdi 7,9,0
	blt- 7,.L20
	li 0,-1
	extsw 9,9
	rldicl 0,0,0,12
	srad 10,0,9
	and. 0,10,11
	beqlr- 0
	lfd 0,.LC0@toc(2)
	lfd 13,.LC1@toc(2)
	fadd 0,1,0
	fcmpu 7,0,13
	bng- 7,.L6
	cmpdi 7,11,0
	blt- 7,.L21
.L16:
	andc 11,11,10
.L6:
	std 11,-16(1)
	lfd 0,-16(1) <-- Load directly into float reg 1 and
	fmr 1,0      <-- avoid subsequent copy.
	blr
.L2:
	cmpdi 7,9,1024
	bnelr+ 7
	fadd 1,1,1
	blr
.L20:
	lfd 0,.LC0@toc(2)
	lfd 13,.LC1@toc(2)
	fadd 0,1,0
	fcmpu 7,0,13
	bng- 7,.L6
	cmpdi 7,0,0
	blt- 7,.L22
	li 11,0
	b .L6
.L21:
	lis 0,0x10
	sldi 0,0,32
	srad 0,0,9
	add 11,11,0
	b .L16
.L22:
	rldicl. 0,11,0,1
	beq- 0,.L6
	lis 11,0xbff0
	sldi 11,11,32
	b .L6



-- 
           Summary: PowerPC - Unnecessary Floating Point Register Copy
           Product: gcc
           Version: 3.5.0
            Status: UNCONFIRMED
          Severity: enhancement
          Priority: P1
         Component: other
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: steinmtz at us dot ibm dot com
                CC: gcc-bugs at gcc dot gnu dot org,steinmtz at us dot ibm
                    dot com
 GCC build triplet: powerpc64-linux
  GCC host triplet: powerpc64-linux
GCC target triplet: powerpc64-linux


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16796

[Bug target/16796] PowerPC - Unnecessary Floating Point Register Copy

[bangerth at dealii dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|other                       |target


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16796

[Bug target/16796] PowerPC - Unnecessary Floating Point Register Copy

[falk at debian dot org]

------- Additional Comments From falk at debian dot org  2004-07-28 18:31 -------
I don't think this is a target problem, but most likely it's just general
register allocator lameness.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16796

[Bug rtl-optimization/16796] PowerPC - Unnecessary Floating Point Register Copy

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|target                      |rtl-optimization
           Keywords|                            |missed-optimization


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16796

[Bug rtl-optimization/16796] PowerPC - Unnecessary Floating Point Register Copy

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-10-08 13:11 -------
Confirmed, the problem is related to the register allocator.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
   Last reconfirmed|0000-00-00 00:00:00         |2004-10-08 13:11:12
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16796

[Bug rtl-optimization/16796] PowerPC - Unnecessary Floating Point Register Copy

[nathan at gcc dot gnu dot org]

------- Additional Comments From nathan at gcc dot gnu dot org  2004-11-10 11:01 -------
The fault is not the register allocator, is is a problem introduced by flow2 and
the existance of conditional returns on PPC.  During register allocation
there is one exit block.  The returns are 'x, x, x+x & x', thus the global
allocator sets the exit block to copy 'x' to the return register.  flow2
replaces some of those branches to the exit block with conditional returns,
because 'x' resides in both DF:33 and DF:32.  The remaining use of the exit
is the load from the union into DF:32, the location of 'x' during the function.



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16796

[Bug rtl-optimization/16796] PowerPC - Unnecessary Floating Point Register Copy

[nathan at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |nathan at gcc dot gnu dot
                   |dot org                     |org
             Status|NEW                         |ASSIGNED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16796

[Bug rtl-optimization/16796] PowerPC - Unnecessary Floating Point Register Copy

[nathan at gcc dot gnu dot org]

------- Additional Comments From nathan at gcc dot gnu dot org  2004-11-11 08:47 -------
2004-11-11  Nathan Sidwell  <nathan@codesourcery.com>

	PR target/16796
	* config/rs6000/rs6000.md: Add DF & SF reg move peepholes.

partial fix. Now we need a better reg-alloc

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
  BugsThisDependsOn|                            |18427
         AssignedTo|nathan at gcc dot gnu dot   |unassigned at gcc dot gnu
                   |org                         |dot org
             Status|ASSIGNED                    |NEW


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16796

[Bug rtl-optimization/16796] PowerPC - Unnecessary Floating Point Register Copy

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2004-11-11 08:49 -------
Subject: Bug 16796

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	nathan@gcc.gnu.org	2004-11-11 08:49:17

Modified files:
	gcc            : ChangeLog 
	gcc/config/rs6000: rs6000.md 
	gcc/testsuite  : ChangeLog 
Added files:
	gcc/testsuite/gcc.dg: ppc-mov-1.c 

Log message:
	.:
	PR target/16796
	* config/rs6000/rs6000.md: Add DF & SF reg move peepholes.
	testsuite:
	PR target/16796
	* gcc.dg/ppc-mov-1.c: New.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.6272&r2=2.6273
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/config/rs6000/rs6000.md.diff?cvsroot=gcc&r1=1.332&r2=1.333
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/ChangeLog.diff?cvsroot=gcc&r1=1.4575&r2=1.4576
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gcc.dg/ppc-mov-1.c.diff?cvsroot=gcc&r1=NONE&r2=1.1



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16796